Challenges for a hybrid CAI-based INS due to trajectory dynamics derived from real inertial datasets

The results of an observability analysis of a CAI-based IMU sensor fusion model, based on findings from an extensive analysis of inertial datasets collected over the last 10 years at the Institut für Erdmessung of the Leibniz University of Hannover, are presented. Datasets are analysed with respect to characteristic peaks occurring in the body frame during acceleration, deceleration, and turn maneuvers. This is done for IMU datasets recorded on board trains and cars. Based on these findings, “characteristic” maneuvers are derived for the forward (x) and right (y) axis of the accelerometer, and the z axis of the gyroscope in the body frame. Maneuvers are derived by ranking multiple possible function fits on a RMSE-based evaluation method. This results in best fitting functions which are used to confirm the observability of different systematic IMU error terms with respect to a CAI-based reference sensor. Turn maneuvers result in dynamics across both accelerometer and the gyroscope axes, which in turn leads to observability of misalignments. For acceleration and deceleration maneuvers, only the longitudinal axis of the vehicle exhibits changes in acceleration, which should also be sufficient to estimate the misalignment terms between the conventional IMU and the CAI-based sensor. Meanwhile, the lever arm (displacement) between the CAI and IMU cannot be reliably estimated by maneuvers considered here, as it requires significant angular rates along two axes. A solution to this problem could be the oscillation due to suspension visible in the car-based datasets, which have a frequency of 0.2-1 Hz, and an amplitude of up to 0.1 rad/s. Based on these results, a follow-up study is suggested with real CAI sensor measurements to estimate the impact of such slow oscillations on the sensor solution.